app.py

from fastapi import FastAPI
from transformers import pipeline
import torch
from pydantic import BaseModel
import os
import numpy as np  # Explicit numpy import

# Fix numpy initialization
np.zeros(1)  # Force numpy load before model


app = FastAPI()

# Disable xformers if needed
torch.backends.cuda.enable_flash_sdp(False)
torch.backends.cuda.enable_mem_efficient_sdp(False)


# Configure paths and device
os.environ["TRANSFORMERS_CACHE"] = "/tmp/huggingface"
os.makedirs(os.environ["TRANSFORMERS_CACHE"], exist_ok=True)

@app.get("/ready")
def readiness_check():
    return {"status": "ready"}


# Load model with error handling
try:
    model = pipeline(
        "text-classification",
        model="win2win/3-epochs-classifier-ver2",
        device="cpu",  # Force CPU for free tier
        device_map="auto",
        torch_dtype=torch.float32,  # Avoid mixed precision
        offline_mode=False,  # Force fresh download
        max_retries=3
    )


    print("Model loaded successfully!")
except Exception as e:
    print(f"Model loading failed: {str(e)}")
    raise

class Request(BaseModel):
    text: str

@app.post("/predict")
async def predict(request: Request):
    return model(request.text)

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)